The paradox runs in a circle: to build an accurate map, the robot needs to know its exact position, but to determine its exact position, the robot needs an accurate map.
Correction Step (Update): The robot compares what it expected to see against what it actually saw. It updates both its own estimated pose and the estimated positions of the landmarks, shrinking the uncertainty cloud for both!
Before scanning a landmark, the robot’s estimated position is a wide, fuzzy cloud of possible locations reflecting high uncertainty. After scanning a known landmark and comparing the expected reading to the actual one, that cloud collapses down to a small, tight region around the robot’s true position, reflecting the sharp drop in uncertainty the correction step provides.
Real-World Example (Autonomous Mine Exploration Rovers): Deep underground mining tunnels have no GPS signals. Autonomous subterranean vehicles enter unexplored caverns without a prior map. Using SLAM, they simultaneously chart 3D tunnel structures while tracking their own trajectory to safely navigate back to the surface.
Real-World Example (AR/VR Headsets): Consumer augmented reality glasses use Visual SLAM (VIO) via onboard cameras to track your head movement in real time while constructing a 3D surface model of your living room furniture.
VEX Robotics Example (Lab 5 Autonomous Mapping): In Lab 5, your VEX robot drives through an unknown maze. As it drives, wheel encoders provide initial movement estimates while distance sensors detect maze walls. The SLAM algorithm continuously merges these noisy streams to produce a clean, accurate arena map while keeping the robot localized.
